1 / 31

[ OP_ Research @ IM.NTU ]

MANET Simulation Studies : The Incredibiles 2005 Mobile Computing and Communications Review, Volume 1, Number 2. Agenda:. Introduction Survey Results Common Simulation Pitfalls Simulation Setup Simulation Execution Output Analysis Publishing Community Resources Conclusion.

tamma
Télécharger la présentation

[ OP_ Research @ IM.NTU ]

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. MANET Simulation Studies : The Incredibiles 2005 Mobile Computing and Communications Review, Volume 1, Number 2 Agenda: • Introduction • Survey Results • Common Simulation Pitfalls • Simulation Setup • Simulation Execution • Output Analysis • Publishing • Community Resources • Conclusion 指導教授：林永松 2008/06/09 D95725009 陳仁偉 [OP_ Research@ IM.NTU]

  2. Author Stuart Kurkowski Education Ph.D. Computer Science, The College of William & Mary, 1993M.S. Computer Science, Michigan State University, 1989B.A. Mathematics, Kalamazoo College, 1987 Michael Colagrosso [ 2 ]

  3. I. Introduction • Mobile Ad Hoc Networks (MANETs) are wireless mobile nodes that cooperatively form a network without Infrastructure • Challenges:Routing, wireless communications issues, resource issues, The leading is simulation • Simulation Study: ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc) from 2000-2005 • MANET simulation studies lack believability [ 3 ]

  4. Focus • This study focused on the following four areas of credibility in research • Repeatable • Unbiased • Rigorous • Statistically sound [ 4 ]

  5. I.A. The Current State of MANET Simulation Studies Repeatability - 1 • MANET research full papers published in MobiHoc • 114 out of the 151 MobiHoc papers published (75.5%) used simulation • Unfortunately, 34 of the 114 published MobiHoc simulation papers (29.8%) did not identify the simulator used in the research. • Network Simulator-2 (NS-2) [34] is the most used simulator in MANET research; 35 of the 80 simulation papers that state the simulator used in the simulation study used NS-2 (43.8%). • No paper made a statement about code availability [ 5 ]

  6. [ 6 ]

  7. I.A. The Current State of MANET Simulation Studies Repeatability -2 • The researcher must identify the simulator and version, the operating system, and all variable settings. • Repeatability is also based on the scenarios evaluated, the techniques used to avoid initialization bias (influence of empty queues, etc., at the start), and the techniques used to analyze the results. • Thus, a published paper must discuss or reference all of these details to meet the repeatability criteria [ 7 ]

  8. I.A. The Current State of MANET Simulation Studies • Unbiased study: • A project must address initialization bias, random number issues, and use a variety of scenarios. • Rigorous study: • Factors such as node density, node footprint, coverage, speed, and transmission range must be set to exercise the protocol under test. [ 8 ]

  9. I.A. The Current State of MANET Simulation Studies • Statistically sound study: • A project must account for initialization bias, execute a number of simulation iterations, provide the confidence levels that exist in the results, and list any statistical assumptions made. [ 9 ]

  10. I.B. Survey Motivation • Simulation type: - terminating or steady-state • 66 of the 114 simulation papers (57.9%) did not mention the type of simulation used in the study. • Pseudo random number generator (PRNG) • The goal is to raise awareness of the issues and to introduce tools to aid MANET researchers in conducting and reporting credible simulation results [ 10 ]

  11. II. Survey Results Research Design • Simulation study questions. • Reviewed each paper individually • For consistency, the same person reviewed all of the papers; to validate the results, we had a second person review all of the papers with a subset of the questions and a third person to correct the few inconsistencies found. • 114 out of the 151 (75.5%) published MobiHoc papers used simulation as the basis for the study. [ 11 ]

  12. [ 12 ]

  13. [ 13 ]

  14. [ 14 ]

  15. III. Common Simulation Pitfalls III.A. Simulation Setup III.A.1. Simulation Type III.A.2. Model Validation & Verification III.A.3. PRNG Validation & Verification III.A.4. Variable Definition III.A.5. Scenario Development III.B. Simulation Execution III.B.1. Setting the PRNG Seed III.B.2. Scenario Initialization III.B.3. Metric Collection III.C. Output Analysis III.C.1. Single Set of Data III.C.2. Statistical Analysis III.C.3. Confidence Intervals III.D. Publishing [ 15 ]

  16. III.A. Simulation Setup • III.A.1. Simulation Type - terminating vs. steady-state • 66 out of the 114 simulation papers (57.9%) did not state • MANET : steady-state • statistically unsound results • III.A.2. Model Validation & Verification • Simulation model • The model must be validated as a baseline to start any experimentation • Not validating the model or verifying code is a common pitfall [ 16 ]

  17. III.A. Simulation Setup • III.A.3. PRNG Validation & Verification • NS-2 Cycle length is 231-1 • Akaroa-2 [11] offers an NS-2 compatible PRNG with a cycle of 2191 − 1. • 2-dimensional request on a PRNG is valid for pproximately 8p3 L, where L is the cycle length. • III.A.4. Variable Definition • There are 538 variables defined in the ns-default.tcl file of NS-2.1b7a and there are 674 variables defined in the ns-default.tcl file of NS-2.27. • Our review of the Tcl driver files show that many simulation driver files leave key parameters undefined. [ 17 ]

  18. III.A. Simulation Setup • III.A.5. Scenario Development • Only 48 of the 109 MANET protocol simulation papers in our survey of published MobiHoc papers provided all three of these input parameters, detailing 61 simulation scenarios. • #36 & #37 • number of nodes : 10 – 30000 • Simulation area : 25m x 25m - 5000m x 5000 m. • Transmission ranges: 3m - 1061 m. • The survey results highlight the wide range of simulation scenarios used to conduct MANET research and the lack of uniform rigorous testing of MANET protocols. [ 18 ]

  19. [ 19 ]

  20. III.A. Simulation Setup • III.A.5. Scenario Development • We validated the wide range of input parameters by comparing the derived parameters of each scenario. • Figure 2 is a scatter plot of all the derived parameters for the 61 sets of input parameters. • The scatter plot reflects the wide range of scenarios and the lack of correlation between parameters. • Figure 2 also shows the lack of independence between parameters, such as node density and node coverage . • the lack of multiple groupings in each plot illustrates that the community is not covering the range of values in a consistent organized manner. [ 20 ]

  21. [ 21 ]

  22. [ 22 ]

  23. III.A. Simulation Setup • III.A.5. Scenario Development • The MANET community needs a way to characterize simulation scenarios in order to evaluate and compare protocols and performance and ensure protocols are rigorously tested. • For example, from Table 2, scenario #8, the simulation area is 3000mx 3000 m, but the transmission range of 1061m lowers the average hop count to only 1.67 hops. This hop count means most source and destination pairs are direct neighbors and the rest have only one intermediate node. [ 23 ]

  24. III.B. Simulation Execution • III.B.1. Setting the PRNG Seed • NS-2 uses a default seed of 12345 for each simulation run • Produce identical results • Negate the independent replication method • None of the 84 simulation papers addressed PRNG issues. [ 24 ]

  25. III.B. Simulation Execution • III.B.2. Scenario Initialization • This pitfall usually occurs from a lack of understanding of the two types of simulation. terminating or steady-state • Only eight of the 114 simulation papers in our MobiHoc survey (7.0%) addressed initialization bias, and all eight use the unreliable method of arbitrarily deleting data. [ 25 ]

  26. III.C. Output Analysis • III.C.1. Single Set of Data • This pitfall is taking the first set of results from a simulation and accepting the results as “truth”. • Only 39 of the 109 MANET protocol simulation papers (35.8%) stated the number of simulation runs executed. • III.C.2. Statistical Analysis • This pitfall concerns not using the correct statistical formulas with the different forms of output. [ 26 ]

  27. III.C. Output Analysis • III.C.3. Confidence Intervals • Confidence intervals are a tool to provide a range where we think the population mean is located relative to the point estimate • 98 of the 112 simulation papers using plots (87.5%) did not show confidence intervals on the plots. [ 27 ]

  28. IV.D. Publishing • The inconsistency prevents the direct comparison of results, limiting research advancements. • A new researcher cannot repeat the studies to start his or her own follow-on research. • When publishing, the authors need to state if the code is available and how to obtain the code. [ 28 ]

  29. IV. Community Resources • The Akaroa-2 suite, which help a researcher monitor simulation execution to determine steady-state and prevent correlation among multiple replications of a simulation. • The interactive NS-2 protocol and environment confirmation tool (iNSpect) , which visualizes the trace file of an NS-2 simulation. The visualizations can be used for scenario development, model validation, protocol verification, and results analysis. • The Simulator for Wireless Ad Hoc Networks (SWAN) , which enables a researcher to create a virtual environment for conducting experiments with MANETs. [ 29 ]

  30. V. Conclusions • Summarizing the four areas of credibility, we found less than 15% of the published MobiHoc papers are repeatable.. • Only seven of the 58 MobiHoc simulation papers that use a public simulator (12.1%) mention the simulator version used. • Only eight of the 114 simulation papers (7.0%) addressed initialization bias and none of the 84 simulation papers addressed random number generator issues. • statistical soundness, 70 of the 109 MANET protocol simulations papers (64.2%) did not identify the number of simulation iterations used, and 98 of the 112 papers that used plots to present simulation results (87.5%) did not include confidence intervals. [ 30 ]

  31. V. Conclusions • Hence, only approximately 12% of the MobiHoc simulation results appear to be based on sound statistical techniques. • This survey of MobiHoc papers showed the current state of MANET research and the lack of consistency, re-enforcing the need for simulation study guidance. • The goals is to make the research repeatable, unbiased, rigorous, and statistically sound [ 31 ]

More Related